Research On Key Technology And System Of Intelligent Information Question Answering

Intelligent question answering is an important application in the field of artificial intelligence.It almost appears in the process of human-computer interaction,such as intelligent customer service and emotional dialogue robots.Users can get the answers to the questions they want through intelligent question answering technology.Especially in the military field,intelligent question answering technology has played a vital role,and it is a key part of the decision-making assistance system.At present,the research work in related fields is mainly divided into retrieval style and generative style.Search question and answer is to use existing knowledge and data to return answers to relevant questions to users by means of retrieval.The generative question and answer uses a pretrained language model to generate corresponding answers through the understanding of the user’s natural language questions.Both methods are more mainstream and considered effective methods.The research of this paper focuses on the research of intelligent question answering technology on military field data.However,data in the military field is difficult to obtain,and the format is irregular.For the question-and-answer system using retrieval type,due to the small amount of data in the military field and the difficulty of manual labeling,it is difficult to achieve efficient and accurate coverage.For generative question answering systems,due to the lack of relevant pre-training models,it is also difficult to generate text answers with strong readability and high confidence.In view of this,this paper proposes an intelligent question answering system that combines retrieval and generative methods,combining the advantages of the two methods.Specifically,this article proposes an intelligent question answering system consisting of three parts.One is a user question classification model,which is used to classify users’ questions and decides to use retrieval or generative methods;the other is a user question matching model,which is used to use retrieval.The user question of the method is matched with the most similar question text in the knowledge base,and the corresponding answer is returned;the third is the user question database query model,which is used to use the user question of the generative method to convert the user question into a SQL database query statement.And generate answer text.First introduce the problem classification method combined with How Net perceptual information.In the task of Chinese text classification,word segmentation and polysemous word ambiguity have always been difficult challenges to overcome.Especially in the military field,once such an error occurs,it may cause irreversible results.To this end,this paper designs a bidirectional lattice long short-term memory network(SK-Lattice)with external knowledge perception capabilities.The network adds a path to link the beginning and end characters of the vocabulary to control the flow of information.At the same time,an additional gating mechanism is designed to give selective attention to the multiple meanings of polysemous words and dynamically emphasize the correct meaning of words.In the experiment,20 classic classification methods were compared on the military field data set and three Chinese text classification benchmark data sets.Due to the small amount of data in the military field,it is difficult to train a problem matching model with high accuracy.For this reason,this paper designs a cross-domain problem matching model X-QR based on adversarial training.X-QR uses standard open source Chinese text data in other fields with sufficient data volume,and uses domain discriminators for adversarial training to achieve the alignment of the source domain data and the target domain verse distribution and the unity of the feature space.In this way,knowledge transfer is carried out,and the problem matching device is used to learn the unified knowledge of the domain,and accurate problem matching is realized on the military field data with a small amount of data.For those who cannot match the existing questions to answer,the generative scheme is adopted.It is worth noting that many data in the military field are stored in relational databases,such as SQL Server.This involves text-to-SQL(Text-to-SQL)technology.Existing mainstream models use a template-based method to generate SQL query statements by filling in slots.However,if the query contains multiple values belonging to different columns,traditional methods may not be able to accurately extract the values.In addition,if the query does not explicitly mention the corresponding column name,it is difficult to infer the correct value.In order to make up for this shortcoming,this paper proposes a new neural network model,namely ER-SQL.ER-SQL uses column content to better extract column characteristics.In addition,ER-SQL uses column representation to learn potential information to reconstruct queries.Finally,in the experiment,the effectiveness of the ER-SQL model was verified using military data and open source data.Finally,this paper also prototyped the intelligent question answering system for the US military’s aerospace equipment,and constructed a complete intelligent question answering system prototype using the designed model algorithm.Starting from the actual needs of the system,intelligent question answering applications in multiple scenarios are realized.Traditional intelligent question answering uses only one technical solution,retrieval or question answering.The characteristics of military data are difficult to overcome by a single technical solution.In order to solve these problems,this paper proposes three important components of the intelligent question answering system for the US military’s aerospace equipment.Combining the advantages of the two technical solutions,the question and answer in the military field can be realized efficiently and accurately.The effectiveness of the algorithm and model is verified in the experiment.At the same time,using a number of technological innovations,a complete intelligent question answering prototype system was designed,laying the foundation for the next step of development.